Deconvolution strategies?

[pfile]

i recently started using a small R-C telescope which (fortunately) came pretty well collimated out of the box.

because i have a pretty crappy mount, almost any length of exposure gives me trailing in RA. i'm autoguiding and in fact without the guider the PE is so bad that the results would be completely unusable.

now, i know the right answer is to get a new mount. but short of that, i've found that motion blur deconvolution can sometimes whip my images into shape.

i've also extracted an (egg shaped) PSF from my images using DynamicPSF. however, deconvolving with such a PSF does not bring the stars back into round. maybe one should not expect them to... or am i doing something wrong?

one question i've had is: is there some difference between doing deconvolution with a line as the PSF and motion blur deconvolution? or is it only a matter of convenience to have multiple methods to generate a PSF?

if i wish to do both motion blur and extracted-PSF deconvolution, i'm assuming that i should do the PSF-based one first followed by the motion blur deconvolution. does that sound correct?

[Nocturnal]

Hi Rob,

take a look at this article I wrote a while ago:

http://deepskystacker.wikispaces.com/Measuring+differential+flexure

even if you do not use DSS for stacking you may be able to grab numbers from your stacking software. I strongly urge you to solve whatever issue is causing your star shapes by improving your capture process.  Once detail is lost no amount of deconv magic can restore it. Crappy mounts can produce decent images with reasonably round stars but you need to know what's broken before you can fix it.

[pfile]

sander - it's the RA motor/gears for sure. if i tell PhD to suppress it's guide output and watch the RA curve, it's a huge sine wave with more than 2 pixels worth of excursion peak-to-peak on the guide camera. i'm nowhere near the territory of flexure here.

when the guiding is running, if i'm balanced properly PhD can sometimes keep the mount to within 1/2 pixel peak-to-peak, which yields reasonable subs at fl=1100mm.

i don't really want to spend the multiple thousands of dollars that's required to get a proper mount... so step 1 is messing around with deconvolution and step 2 is possibly to disassemble the mount and try to hypertune it (although there's no 'hyper' here, it's just trying to make it work properly).

it will be interesting to see if meade's new mount is any good. but you get what you pay for, and it's just less than $1k.

[Nocturnal]

Hi Rob,

Every reasonable mount has a sine wave if you do not guide it so I'm not sure why you think that's a problem. My EQ6 had a pretty severe PE which PHD guided away with good success.

I'll just tell you that when I went from an EQ6 ($1500) to an EM-400 ($10k) my images didn't magically get better. Even an EM-400 has a few arc seconds worth of PE and requires guiding.

Show me your stacked image graph as suggested in my article please. I'll send you $5 via paypal if your movement line is flat, I kid you not. I dare you. If you do not have the means to do the test you can upload the first and the last image in a session, jpgs is fine. I will do the measurement for you and still pay you $5 if have zero image shift. If you do have image shift you owe me nothing and you'll have the root cause of your imaging issues.

The most expensive mount in the world does not prevent differential flexure if it isn't guided correctly. Your modest mount may well be capable of producing round stars with some adjustments to your guide setup.

[pfile]

there is definitely drift. it is in the RA direction. i have measured this.

[Nocturnal]

Ok so then you have differential flexure which is not a property of your mount, nor of PHD tuning but of your guide scope setup. That's good news as it means you may be able to improve the situation without getting a more expensive mount. If you post a picture of your imaging and guide scopes and how they are connected perhaps we can give some suggestion.

Getting rid of DF is near impossible but you should be able to reduce it to the point where you can take subs that are long enough to be useful while keeping stars round enough. When I do hyperstar imaging I have to stay below 4 minute exposures to keep my stars round enough. When I use an OAG for 2800mm FL imaging I can go much longer.

[pfile]

i can do that and it may be the solution but quite honestly i am interested in the math and the application of different deconvolution algorithms as well, which is really why i posted the question.

[Nocturnal]

Oh, ok.

[pfile]

btw you are absolutely correct about the DF, the dithering obfuscates things a bit but i'm definitely movin' right along in RA (dec guiding is off because the mount does crazy things, like sticking, in DEC. fortunately i can drift align the dec axis quite well.)

my guide camera mounting is very poor. i need to put it on a rail; it's currently in a set of spotting scope rings. perhaps i can improve that mechanically but i think the rail solution would be better.

after i get my balance problems sorted out i'll start working on the DF. i have some counterweights due to arrive tomorrow.

[Nocturnal]

If you like to bounce some ideas of me/us when you get around to fiddling with your guide scope please feel free to post here or on the DSS forums or send me a note. As you've probably noticed DF is a hot button issue for me and I'll gladly help others diagnose and improve it.

Sorry, I don't understand deconv enough to help other than saying that with the star analysis tool you should be able to create a representative PSF for input into the deconv. There is a forum thread about that somewhere here.

[pfile]

i *think* this is probably solved. i did another run last night after properly balancing the tube and got an RA drift of 1.8 pixels over about 54 minutes, or about 0.033333px/min.

DEC turned out really bad - 1 pixel per minute. unfortunately last night before running the DF test, i decided to refine my drift alignment with WCS and accidentally swapped the meaning of the red and green circles when adjusting the mount, thus worsening it. before starting up again, i re-did the drift alignment with 1200s of drift time and my overnight run had 0.3px/min DEC drift. this is all at a FL of 1100mm with a canon 50d at prime focus (APS-C sensor with 4.7um pixels)

i can't say too much about the RA on the overnight run, since i turned dithering on. assuming the dithering was completely random then RA drift was very minimal, but i'm pretty sure the final RA position is biased. i've noticed that sometimes PhD claims to have dithered and has not actually done it.

anyway besides balancing properly i tried to tidy up the guide camera USB cable and really tighten down the screws in the ring mount. since this was not scientific at all, i can't say if the problem was balance or DF.  i didn't do much to the 3 cables going to the camera, and cable drag there could contribute to DF as well...

[pfile]

well... it's not totally solved. since balancing properly my stars are a bit rounder.

but, the elongation is still there. i've found that it's relatively invariant across exposure time. for instance, DPSF says r is in the neighborhood of 0.75-0.68 for exposures of 60s, 120s and 240s. at 720s r is worse but now i might be looking at some DF in the mix at that point.

i finally solved a field from my guide camera and found that my guider scale is 8.9 arcseconds/pixel. my imaging camera is about 0.88 arcseconds/pixel. now, i know that PhD does manage subpixel accuracy, so a mismatch like this in and of itself is probably not fatal. in a typical run, i see 0.15 pixels error in RA (RMS). if my math is right, this translates to 0.42 guider pixels peak-to-peak error, and that translates to roughly 4 pixels on the imaging camera, which looks to be in the ballpark.

i was about to conclude that my guide camera focal length is too short, but... i'm not sure what fraction of a pixel PhD can actually resolve with this camera, but assuming it's 0.1 pixel, that far exceeds the best seeing that i'd expect to see around here. so it would seem that my guider focal length is not too short (currently it's 250mm).

so... not sure what to do next here.

i do know for a fact that my mount elevation is not spot-on. i have been assuming that the mount PE error far exceeds any polar misalignment in RA, and so the bad alignment would be a 2nd order effect. but now i'm starting to wonder about that.

[Nocturnal]

Star elongation still only in RA?

I would attach a webcam or something similar to your imaging camera and record a few minutes of star movement while guiding. This will show you how your imaging scope moves. Use perecorder or metaguide together with pecprep if you don't have software for this yet.

[pfile]

yeah the elongation direction is RA. last night i was making some very short exposures of the trap (like 15s) and noticed the same kind of elongation, though that may have been caused by mirror slap. and then this morning while taking flats i noticed my rail-mounted counterweight was loose, so last night's results might all be bunk.

funny you should mention the webcam - i had the same idea, so i used BackyardEOS's planetary mode on one of the bright stars in the pleiades last night. i have a roughly 400s movie. but i was wondering how to analyze it. so metaguide or pecprep, huh?

[Nocturnal]

You could try running the movie through registax if you enjoy such pain. My strategy would be to have MG or perecorder look at your webcam -as you record- and have pecprep analyze the resulting deflection file.